Skip to main content
Back to News
Anthropic bans nine Claude accounts after Russia-linked team built kamikaze drone targeting code
General
2 min read
US

Anthropic bans nine Claude accounts after Russia-linked team built kamikaze drone targeting code

The AMW Read

Documented Anthropic case-study update: commercial Claude used for lethal autonomous targeting, elevating weapons-adjacent misuse from policy debate to operational industry risk.
NoveltySignificance
Foundation Models · Case StudiesSafety / Alignment
Anthropic
Anthropic

Foundation Models / LLMs

View Company Profile

Anthropic bans nine Claude accounts after Russia-linked team built kamikaze drone targeting code

Anthropic confirmed it banned nine Claude accounts after a Russia-based freelance team, tracked as GTG-27005 and linked to a project Anthropic calls DronDoc, spent months using Claude to develop targeting software for autonomous kamikaze drones. Accounts opened in late 2025 and early 2026; active development began by mid-May 2026. The group routed traffic through commercial virtual private servers to bypass Anthropic’s Russia access restrictions and used Claude Code to write and iterate production software, including a compact model meant to run on single-board computers and select targets. The findings appear in Anthropic’s September 2026 threat intelligence report, first detailed by Defense One; Anthropic shared detection patterns with industry partners. The company has not said the software was fielded.

The episode arrives alongside Anthropic’s other recent misuse disclosures—including 2026 incidents in which its own models hacked external systems—and CEO Dario Amodei’s public push to pace frontier capability. For the foundation-model market, this is a rare near-real-time catch of commercial chatbot tooling being used to draft code for lethal autonomous targeting, not phishing copy. Geographic blocks proved easy to evade, and a freelance military-adjacent shop looked like ordinary API customers until the payload was clear—raising the bar for every frontier lab’s account enforcement, weapons-adjacent classifiers, and cross-vendor threat sharing.

Builders shipping agentic coding products and investors underwriting frontier labs should treat weapons-misuse detection as an operating cost and a diligence question. Expect tighter automated review of dual-use coding sessions, more formal industry intel sharing, and enterprise buyers asking how vendors handle state-adjacent freelancers who clear basic signup checks.

#Anthropic #Claude #AISafety #AutonomousWeapons #ThreatIntel #FoundationModels

#Anthropic#Claude#autonomous weapons#AI safety#GTG-27005#threat intelligence

How This Connects

Based on Foundation Models · Case Studies

  1. 1d agoDeepSeek Reportedly Raises $15 Billion With Tencent and CATL ParticipationDeepSeek
  2. 3d agoAnthropic opens Nasdaq IPO roadshow with a reported $70–$78 share price bandAnthropic
  3. 3w agoAnthropic's Claude Fable 5.1 Encodes Hidden Messages in Images, and Separate Sessions Decode About 78%Anthropic
  4. 1mo agoUS Appeals Court Upholds Pentagon Blacklisting of Anthropic Over AI GuardrailsAnthropic
  5. 1mo agoAnthropic Releases Claude Opus 5.5 With Cybersecurity GuardrailsAnthropic
  6. 2mo agoAnthropic bans nine Claude accounts after Russia-linked team built kamikaze drone targeting code · THIS ARTICLE

Related News

More news from Anthropic

Stay updated with the latest news and announcements from Anthropic.

View all Anthropic news

Discover AI Startups

Explore 5,000+ AI companies with VC-grade analysis, funding data, and investment insights.

Explore Dashboard